You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Higher-order agents that build, optimize, and supervise other agents. Automated design of agent harnesses, optimization of compound agentic systems (DSPy, GEPA, Trace), self-improvement, evaluation/benchmarks, misalignment/safety, governance.
Verification as a first-class problem for autonomous agents. Safety/robustness of verification, environment-grounded verification and simulators, diverse verifiable signals, meta-agents, self-evolving agents, scalable oversight.
First NeurIPS workshop on interpreting agent behavior. Human-centered interpretation for understanding AI agents, human-agent interaction, agent decision-making transparency.
Agent harnesses and multi-agent systems for scientific discovery. Generalist vs. specialist models, autonomous lab-in-the-loop integration with robotics for closed-loop experimentation.
Interaction between pre-training and post-training in LLMs. What base models need for post-training success, roles of SFT/preference optimization/RL/self-improvement/distillation.
First NeurIPS workshop on diffusion language models. Foundations (theory, scaling), efficiency (parallel decoding, fast sampling), reasoning (self-correction, test-time scaling, planning).
Scaling FM training across distributed, heterogeneous compute — inter-datacenter to internet-scale. Communication-efficient optimization, fault-tolerant distributed systems, security in decentralized training.
Linguistic/symbolic medium as a first-class design axis for FMs. Language representation design, compositionality/semantics, cross-linguistic typology, mechanistic interpretability of linguistic structure.
Foundation models for temporal systems. Forecasting/simulation, multimodal temporal data, temporal models (state-space, generative), evaluation/reliability under distribution shift.
AI for drug discovery: closing the gap between benchmarks and real-world impact. Prospective validation, learning under scarce data, AI co-scientists, trustworthy decision-aware AI.
ML for molecular simulation. Learned potentials/force fields, differentiable molecular dynamics, conformational ensembles, molecular foundation models, agentic ML for MD/QM.
Power systems as a first-class ML challenge. Evaluating methods under realistic operating conditions with physics constraints, real-time closed-loop operation.
Stress-testing AI for biology. OOD generalization, causal vs. spurious correlation, limits of foundation models for biology. Focus on failure analysis and negative results.
Neuro-symbolic embodied intelligence. Integration of perception, RL, LLMs, planning, and symbolic knowledge in agents acting in physical/simulated environments.
Attributing model behavior to training data. Contributive attribution, corroborative attribution, downstream applications in law/journalism, tracing synthetic data.
ML for computer systems — replacing heuristics with learning for compilers, databases, memory management, chip design. LLMs and agentic workflows for systems challenges.
Africa in AI: Scaling African AI for global impact. Biomedical AI, climate resilience, education, language technologies, clean energy, data sovereignty.
Foundations of Agentic Systems Theory. Emergent behavior in systems of LLM-based agents using theory from complex systems, developmental biology, organizational sociology, and cognitive science. Theory of mind, norm formation, compositional safety/governance.
Small language models as the foundation of agentic AI on resource-constrained devices. Compression/distillation, hardware co-design, on-device evaluation benchmarks, privacy-preserving local processing.
Post-training adaptation of robot foundation models. Human feedback, policy optimization, world models, reasoning/memory, cross-embodiment adaptation, evaluation and safety.
Foundations of efficient deep learning. Sparsity, compression, efficient FMs, compute-optimal training, theoretical limits of efficient AI under resource constraints.
Foundations of LLM Post-Training in changing environments. Statistical identifiability, optimization dynamics, robustness to misspecification, adaptation vs. capability preservation.
Neural network weights, gradients, representations as a data modality. Model merging, meta-learning, mechanistic interpretability, neural architecture search, AI supply chain analysis.
RL for adaptive experimentation and real-world scientific discovery. RL for autonomous labs, domain randomization, safe exploration, field experiments.
Agentic LLM systems as co-scientists in molecular sciences. Representation learning, FMs for chemistry/biology/materials, closed-loop self-driving labs.
ML with imperfect scientific simulators. Simulation-based inference, model misspecification diagnostics, surrogate/emulator modeling, multi-fidelity modeling.
ML for spatial transcriptomics, spatial proteomics, and multiplex imaging. Spatial representation learning, cell-cell communication, foundation models for spatial biology.
Representation learning in physical systems. Self-supervision, transfer learning, sampling, tokenization for scientific data respecting conservation laws and geometry.
Privacy in the era of large opaque models. Memorization and privacy risks when FMs/LLMs are trained on sensitive data. Brings together privacy, ML, HCI, law communities.
Trustworthy AI for Good. AI safety, social good, and policy/governance. Model auditing, red-teaming, disinformation detection, cooperative AI. Special track on multi-agent security.
Geometric Distributional Deep Learning. Learning systems that jointly model geometry and distributional nature of data. Optimal transport, generative modeling, graph/manifold learning.
E-values, anytime-valid inference, sequential testing, game-theoretic probability with modern ML. Safe testing, conformal prediction, applications to LLM evaluation and auditing.
ML for semiconductor design. Physical design, logic synthesis, GNNs for circuits, LLMs/foundation models for HDL, agentic systems for chip design workflows.
Resource-aware agentic AI. Agents that reason about and adapt to resource budgets (compute, energy, memory, latency). Resource-aware planning, on-device deployment.
6th Mathematical Reasoning and AI. Agentic AI and math — building agents that conjecture, formalize, prove, verify, collaborate with human mathematicians.
Interpretability as a tool for scientific discovery. Turning learned representations into novel, testable knowledge across protein structure, clinical forecasting, astronomy, climate.
Medical reasoning with vision-language FMs. Interpretable chain-of-thought diagnosis, multimodal grounding, probabilistic reasoning for clinical decisions.
AI agents for biomedical imaging and multimodal clinical data. Tool-using systems for multi-step analyses, evaluation of reasoning/uncertainty/error recovery.
World models for high-stakes health. Patient world models for clinical trial simulation, counterfactual outcome prediction, treatment effect modelling, causal inference.
AI & Science: Evolution or Extinction? Scientific integrity in the era of human-AI collaboration. Epistemic values, evaluations, sociotechnical guardrails.
Protecting children in AI ecosystems. Safe data curation, robust model design (jailbreak resistance, unlearning), deployment safeguards, age-appropriate AI.
ML x Operations Research. Integrating Generative AI into decision sciences — policy design with digital twins, agentic AI for autonomous operations, healthcare/logistics/finance.
Physical World AI. How AI perceives, represents, and reasons about the physical world. Physical geometry (3D/4D), physical characteristics (materials, affordances), physical sensors (tactile, audio, depth).
AI research and perspectives from the Global South.
Contributing
Found an error or want to add more details? Open an issue or submit a PR!
License
This list is provided for informational purposes. Each workshop has its own terms — please visit their official websites for the most up-to-date information.
Acknowledgments
Data sourced from OpenReview and individual workshop websites.
About
Workshop list for NeurIPS 2026, as I didn't find it anywhere else.